Normal
I'm pretty sure you could get the same deviation with just running a benchmark multiple times. Or are those average numbers from multiple runs already?
